The Last Mile Mindset: Why Finishing Beats Starting in the AI Era
· Nia
The Last Mile Mindset: Why Finishing Beats Starting in the AI Era
There's a phrase making the rounds in venture capital right now that captures something profound about the moment we're in. Tiffany Luck, a partner at New Enterprise Associates, calls it "the last mile problem."
Here's the setup: AI tools like Claude and ChatGPT can take any knowledge worker from 0% to 80% on nearly any task. Research a topic? Done. Draft an email? Easy. Generate a first pass at code? Trivial. But that final 20% — the part where work becomes finished, where it's production-ready, where it actually ships — that's where 99% of people are still doing everything manually.
"For 99% of people, AI isn't yet running in the background while their hands are off the keyboard," Luck told Crunchbase News recently.
This observation isn't just about technology. It's about mindset.
The 80% Trap
We are living through the greatest explosion of starting in human history. More projects initiated, more documents drafted, more code generated, more ideas explored than ever before. AI has made starting essentially free.
And that's exactly the problem.
When starting is free, the bottleneck shifts entirely to finishing. The competitive moat moves from "who can think of this" to "who can ship this." From "who had the idea" to "who actually delivered the artifact."
I see this everywhere:
- Founders with 47 AI-generated pitch decks who haven't talked to a single customer
- Developers with repositories full of half-built prototypes, each started with an AI coding assistant
- Writers with folders of AI-drafted outlines that never became published pieces
- Companies with elaborate AI strategies that never produced a single deployed workflow
The 80% trap is real, and it's getting worse as AI makes the initial 80% faster and easier. The gap between "started" and "shipped" is widening.
What BBVA Learned About Shadow AI
Harvard Business Review recently published a fascinating case study about BBVA, one of Europe's largest banks. Their corporate AI program was struggling — clunky tools, slow rollouts, unimpressive results. Sound familiar?
But underneath the official program, a shadow revolution was happening. Employees were quietly using personal ChatGPT and Claude accounts on the side, often without telling IT or compliance. One official at a large central bank reported that employees working on secure, no-AI bank PCs kept personal laptops open to their favorite LLM homepage.
BBVA's breakthrough? They stopped fighting this behavior and instead "followed employees' lead on adoption." They ditched centralized mandates and let the people doing the actual work determine how AI should integrate.
The mindset shift here is critical: top-down AI strategies fail because they optimize for control. Bottom-up AI adoption succeeds because it optimizes for finishing actual work.
The employees weren't using shadow AI for fun. They were using it because the official tools didn't help them close the last mile of their daily work.
The Finishing Mindset
So what does a "last mile mindset" actually look like in practice? Here's what I've observed from the builders who are actually shipping:
1. They Define "Done" Before They Start
AI makes it seductively easy to explore. You can generate ten variations, explore fifteen angles, research twenty competitors — all in an hour. But exploration without a clear definition of "done" is just productive-feeling procrastination.
The finishers I know start every session with a concrete deliverable: "I will have a deployed feature by end of day." Not "I will explore the feature space." Not "I will research best practices." A specific, shippable artifact.
2. They Resist the Refinement Loop
Here's a trap unique to the AI era: because iteration is cheap, you can endlessly refine. Ask the AI to make it better. Then again. Then again. Each version is marginally improved, but you never actually ship.
The finishers set a "good enough" threshold in advance and stick to it. They know that a shipped 85% solution beats an unshipped 97% solution every single day.
3. They Automate the Last Mile, Not Just the First Mile
Most people use AI to generate first drafts. The highest-leverage use is actually automating the finishing work — the formatting, the deployment, the distribution, the testing, the follow-up.
This is exactly what vertical AI companies like Harvey (legal) and Samaya AI (equity research) are doing. Their moat isn't generating text — any LLM can do that. Their moat is producing finished artifacts that look exactly like what a team of analysts would deliver. The output is indistinguishable from done human work.
4. They Embrace Imperfect Shipping
There's a psychological phenomenon where AI-assisted work feels less "yours," which paradoxically makes you more protective of it. You want to refine it until it feels authentically authored. This is a mindset trap.
The best builders I know have made peace with the fact that their workflow is collaborative with machines. They don't over-polish to prove they didn't use AI. They ship fast and iterate based on real feedback.
The Organizational Implications
This isn't just a personal productivity insight — it's reshaping how entire organizations need to think.
The companies winning right now are the ones that:
- Measure outputs, not inputs. It doesn't matter how many hours someone spent or how many tools they used. Did the work ship?
- Remove finishing friction. Every approval layer, review cycle, and deployment gate that adds time between "done" and "live" is now a competitive liability.
- Empower individual judgment. BBVA's lesson is clear — centralized control over AI adoption slows everything down. Let people closest to the work decide how to finish it faster.
- Invest in last-mile tooling. The sexiest AI investment isn't in generation — it's in deployment, testing, formatting, and delivery automation.
A Personal Reckoning
Here's what I've had to confront in my own work: AI made me a better starter long before it made me a better finisher. I had to consciously develop the discipline to close loops, ship imperfect work, and resist the endless refinement that cheap iteration enables.
The mental model that helped me most: treat every AI interaction like it has a shipping deadline. Not "let me explore this," but "let me finish this in the next 30 minutes."
This constraint — artificial urgency applied to AI-assisted work — is the closest thing I've found to a reliable antidote for the 80% trap.
The New Competitive Advantage
In a world where everyone has access to the same AI models, the same starting capabilities, the same ability to go from 0 to 80% in minutes — the differentiator is purely psychological.
It's the willingness to ship imperfect work. The discipline to define "done" narrowly. The organizational courage to remove finishing friction. The personal mindset that values a shipped 85% over an unshipped 95%.
The last mile isn't a technology problem. It's a mindset problem. And the founders, builders, and organizations that solve it will own the next decade.
Start less. Finish more. Ship today.